A neural network-based ECMS for optimized energy management of plug-in hybrid electric vehicles

Abstract

For plug-in hybrid electric vehicles, the equivalent consumption minimum strategy is typically regarded as a battery state of charge reference tracking method. Thus, the corresponding control performance is strongly dependent on the quality of state of charge reference generation. This paper proposes an intelligent equivalent consumption minimum strategy based on dual neural networks and a novel equivalent factor correction, which can adaptively regulate the equivalent factor to achieve the near-optimal fuel economy without the support of the state of charge reference. The Bayesian regularization neural network is constructed to predict the near-optimal equivalent factor online, while the backpropagation neural network is designed to forecast the engine on/off with the aim of improving the quality of equivalent factor prediction. The corresponding neural network training takes advantage of the global optimality of dynamic programming. Besides, the novel equivalent factor correction can guarantee that the electrical energy is gradually consumed along the trip and the terminal battery state of charge satisfies the preset constraints. A series of virtual simulations under a total of nine driving cycles demonstrates that the proposed method can deliver a competitive fuel economy comparing to the optimal solution derived from the dynamic programming, as well as regulating the battery state of charge to reach the desired terminal value at the end of the trip.